Cloud-run agents in Antigravity start reaching Ultra and Pro

Google’s November 3 release of the Antigravity App, v2.29.0, starts rolling out remote agents to Google AI Ultra and Pro subscribers. With it, conversations in a cloud project run in the cloud rather than on the user’s own computer. Voice mode, built on Gemini Live, arrives in the same release. Both are gradual rollouts, so an up-to-date app may not show either one yet.
That caveat comes from Google’s Antigravity changelog. It also says any new version can take a few days to reach all users, which leaves two separate waits: one for the 2.29.0 build itself, and one for the two features that are still being switched on. Remote agents also have a browser entry point: cloud conversations open at antigravity.google.com/remote as well as in the app. Voice mode starts from a button in an empty message box and hands spoken requests to the agent.
The changes apply to the Antigravity App, the standalone desktop app for managing agents that Google added with Antigravity 2.0 in May. The original editor is still a separate product, the Antigravity IDE, and the CLI and SDK have their own changelog entries.
Of the 22 improvements in the release, only voice mode and remote agents carry a rollout warning. The rest ship with the update, and a few of them change what you see on first launch:
- Jupyter
.ipynbfiles open natively in the file viewer, with code cells, Markdown, tables, images and cell outputs rendered. - The built-in browser saves downloads straight to the downloads folder without blocking prompts, and its address bar suggests pages from browsing history. The default download folder and clearing history are both under Settings > General.
- Unless you picked your own theme and font, the app switches to the Onyx theme and Google Sans. Both can be changed under Settings > Appearance, which also gains a Fill width option and a Screen Reader Mode toggle.
- Keyboard shortcuts can now be customized under Settings > Shortcuts.
- The Automations page splits automations into Scheduled, Always on and Paused lists, with Next run and Last run times.
- The agent picker has moved into the model menu, and single-repo projects pick the branch and Git worktree from one menu.
Several of the 13 fixes affect agent work directly. Git worktrees created for subagents now carry over uncommitted local changes. Switching to a deleted Git branch could overwrite a file with the same name, and that is fixed too. Subagent rows and side-panel tabs had stopped updating their working or waiting status, and messages sent while the agent was idle could end up in the queue instead of starting a turn. Sign in with Google redirects and OAuth popups inside the built-in browser had also been failing, and now work.
The fix list also covers the background drain: a conversation database watcher could loop repeatedly and drive CPU usage high, and 2.29.0 fixes it.
Related stories
- Gemini 4 Argon reportedly lands in Antigravity for Pro users
- OpenAI opt-out skips personal Daybreak Blue sessions
- GitHub Copilot gets five new features at Universe’26
- Google opens Android Studio and CLI to Claude Code and Codex
- OpenAI says its Codex app should eat 60% less disk space
- Google Play now runs vibe-coded apps without a dev account
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.
